This invention relates generally to a method and apparatus for switching and grooming data units in a communications network in a timely manner while providing low switching complexity and performance guarantees, over a plurality of communications links with a plurality of transmission rates.
Circuit-switching networks are still the main carrier for real-time traffic, are designed for telephony service, and cannot be easily enhanced to support multiple services and carry multimedia traffic in their native packet formats. Circuit-switching is based on extremely accurate clock frequencies needed for byte-by-byte switching. This enables circuit-switching networks to transport data streams at constant rates with a small delay jitter. Finally, the clock accuracy for SONET requires a heightened amount of accuracy as line transmission speed increases.
Due to the statistical multiplexing of packet streams, packet switching networks handle bursty data more efficiently than do circuit-switching networks. However, current packet switches and routers operate asynchronously, and only provide “best effort” service in which end-to-end delay and jitter can be neither guaranteed nor bounded. Furthermore, statistical variations of traffic intensity in packet switching networks often lead to congestion that results in excessive delays and loss of packets, thereby significantly reducing the fidelity of real-time streams at their respective points of reception. Finally, since current packet switches and routers electronically process the header of each packet to be routed and switched, a high processing power is required and the scalability of the packet switching network is limited.
In circuit switches routing is time-based. A time period, called frame, is divided into very small slices of time, called slots, that contain only one byte of information. The absolute position of each time slice within each time period determines where its respective byte is routed.
Time-driven switching/routing, in accordance with some aspects of the present invention, supports more sophisticated and flexible timing than circuit switching. Consequently, time-driven switching provides better support of video-based multimedia applications. In accordance with the present invention, the size of time frames used for time-driven switching is larger than the size of time slots used in circuit switching; consequently, time-driven switching is far less complicated than circuit switching. In accordance with the present invention, time-driven routing is based on control information included in at least one of the headers and trailers of selected ones of the time frames, which current circuit switching cannot provide for.
In accordance with the present invention, time-driven switching/routing uses a Common Time Reference (CTR). The CTR, though used in time-driven switching, is not used in circuit switching. The above discrepancy between time-driven switching and circuit switching has far reaching implications during comparison of the two. For example, use of a CTR in a network deterministically ensures that time frames as define by different nodes do not slip, thus enabling the deterministic pipeline forwarding of time frames. The above assurances of a CTR exist in contrast to the properties of a circuit switching network, where (1) there are time slot slips, and (2) deterministic pipeline forwarding is not possible.
In U.S. Pat. No. 5,418,779 Yemini et al. disclose a switched network architecture that uses time. Time is used in order to determine when a plurality of switches can transmit over a predefined routing tree to one destination. This kind of tree is known as “sink” tree since the destination switch functions as a “sink” for the transmission from all switches. The time interval in which the plurality of switches transmits to a selected “sink” destination switch is called time band. In different time bands the plurality of switches are transmitting to a different single “sink” destination switch. Network switches change their configuration between time bands in order to build the proper “sink” tree during each time band. The present invention does use neither “sink” trees nor time bands for transmission over “sink” trees.
Yemini's invention may not be realizable in communications networks with end-to-end propagation delays that are not much smaller than the time band durations. In general, in Yemini's invention the end-to-end propagation delays introduce a non-trivial scheduling problem that may or may not have a solution. Furthermore, Yemini's invention does not discuss or specify how to take into consideration the link propagation delays and the end-to-end propagation delays. Consequently, general topology switched network cannot be built the way it is taught by Yemini's et al. invention.
Yemini's invention has another problem, which is congestion, that is the direct result of using “sink” trees. Data units received from different upstream switches contend for a single outgoing link towards the root of the “sink” tree. The present invention does not have any congestion. This is a direct consequence of using in the current invention completely different system operation principles and methods.
For example, in Yemini's et al. patent there is no pipeline forwarding: data units do not proceed in a lock-step fashion through the communications network, as it is the case in the present invention. The lack of pipeline forwarding leads to the above mentioned scheduling and congestion problems. Such problems are due to the fact that incoming time bands of Yemini's invention are not aligned in different input ports of the network's switches. Furthermore, it was not specified what are the temporal relationship of the same and different time bands on different “sink” tree switches when the link propagation delay and the end-to-end propagation delay are not zero. In contrast, time frames in the present invention are aligned with a Common Time Reference (CTR) on every switch.
Optical data communications using a single wavelength consist of the transduction of a single data stream into a series of pulses of light that are carried over an optical fiber. These pulses of light are of a single wavelength. Use of a single wavelength vastly under-utilizes the capacity of an optical fiber, which is capable of carrying a large number of signals each at a unique wavelength. Due to the nature of propagation of light signals, an optical fiber can carry multiple wavelengths simultaneously. The process of carrying multiple discrete signals via separate wavelengths of light on the same optical fiber is known in the art as wavelength division multiplexing (WDM). Many optical components, including, but not limited to, WDM multiplexers, WDM de-multiplexers, star couplers, tunable lasers, filters, waveguide grating routers (WGRs) are deployed in optical networks featuring WDM, and consequently used in the embodiments presented in this disclosure. [T. E. Stern and K. Bala, “Multiwavelength Optical Networks: a Layered Approach,” Prentice Hall PTR, Upper Saddle River, N.J., USA, ISBN 020130967X. R. Ramaswami and K. N. Sivarajan, “Optical Networks: a Practical Perspective,” Morgan Kaufmann Publishers, San Francisco, Calif., USA, ISBN 1-55860-445-6. H. J. R. Dutton, “Understanding Optical Communications,” Prentice Hall PTR, Upper Saddle River, N.J., USA, ISBN 0-13-020141-3].
The present invention contains the novel combination of: (1) time-based switching and routing and (2) WDM technology. WDM includes the capabilities for (1) dynamic tunable wavelength transmission, (2) dynamic and static wavelength switching, and (3) tunable wavelength reception.
The increasing demand for communications capacity has led to general deployment of Wavelength Division Multiplexing (WDM), which requires extremely high capacity switches. Lambda or static wavelength switches address this need for extremely high switching capacity by switching a whole wavelength from an input optical fiber link to an output optical fiber link without requiring any processing of the transmitted data units. In accordance with the present invention, WDM with whole lambda_switching will be deployed in the network's optical core. However, switching of whole lambdas (e.g., lambdas of OC-192) is inefficient and costly because of three inherent problems it causes:                1. N square problem: the number of lambdas needed to accommodate all the possible connections among all access points is on the order of the square of the number of such access points. This will limit the size of the optical core.        2. Bandwidth mismatch problem: there is a substantial bandwidth mismatch when extremely high capacity backbone networks feed low capacity access links. As data leaves the core and is moved by packet switches towards the edge, buffers at access links frequently become congested, causing increased delays and dropped packets.        3. Traffic unbalancing problem: the traffic load across the network is not evenly distributed, i.e., it is not balanced. Thus, trying to satisfy the traffic load requirements using whole lambda_switching remains both inflexible and inefficient.        
Adding the capability of switching_fractions oflambdas, or Fractional Lambda Pipes (FLPs), solves the above three problems. This approach, also called Fractional Lambda Switching (FLSw), will permit the optical core to be extended far closer to the network edges while reaching the lower speed network access devices with a bandwidth that matches their operation capability.
FLSw dynamically switches lambda fractions while carrying data units (e.g., IP data packets, and SONET STS1 frames), in a heterogeneous (mix of very high speed and very low speed links) meshed network, while providing deterministic performance guarantees. The size of fractional lambda pipes can be dynamically allocated to satisfy the specific needs of the access networks to which a fractional lambda pipe is connected. Small capacity FLPs can be used at the periphery to access low speed sub-networks, such as, cable modems, xDSL, VoIP gateways and wireless.
Fractional Lambda Switching (FLSw) combines the advantages of circuit switching and packet switching. FLSw is used for constructing a Fractional Lambda Pipe (FLP). A FLP is equivalent to a leased line in circuit switching. A FLP is realized by two simple elements:                1. A Common Time Reference (CTR™) throughout the network that is globally aligned with Coordinated Universal Time (UTC); and        2. Pipeline Forwarding (PF™) of time frames (logical containers of data packets) across FLPs.        
The CTR is a reference clock used to realize pipeline forwarding of time frames, both within switches and across FLPs. The CTR™ is received via either the Global Positioning System (GPS), which is globally available at a low cost with an accuracy of 10–20 nanoseconds, or any other UTC (Coordinated Universal Time) distribution system such as the GLONASS (Russia) and, in the future, Galileo (Europe).
The common time reference or more specifically, the UTC second, is partitioned into time frames. The duration of a time frame is a link parameter—fast links might use shorter time frames, while slow links might use longer time frames. Contiguous time frames are grouped into time cycles, and contiguous time cycles are grouped together into contiguous super cycles. The duration of a super cycle is one UTC second, as shown in FIG. 2, and both the duration of time frames and the number of time frames in a cycle can be chosen for convenience. For example, a 1 Gb/s link might use time cycles of 100 time frames, each time frame having a duration of 125 μs; while a 10 Gb/s link might use time cycles of 1000 time frames, each time frame having a duration of 12.5 microseconds. For both links, each time frame will carry the same 15,625-byte payload, and there will be 80 time cycles in each super cycle or one UTC second, as shown in FIG. 2.
The common time reference can be realized by using UTC (Coordinated Universal Time), which is globally available via, for example GPS (Global Positioning System). By international agreement, UTC is the same all over the world. UTC is the scientific name for what is commonly called GMT (Greenwich Mean Time), the time at the 0 (root) line of longitude at Greenwich, England. In 1967, an international agreement established the length of a second as the duration of 9,192,631,770 oscillations of the cesium atom. The adoption of the atomic second led to the coordination of clocks around the world and the establishment of UTC in 1972. The Time and Frequency Division of the National Institute of Standards and Technologies (NIST) (see http://www.boulder.nist.gov/timefreq) is responsible for coordinating UTC with the International Bureau of Weights and Measures (BIPM) in Paris.
FIG. 3 shows an example of the pipeline forwarding of time frames—i.e., the data units transmitted and received during a time frame—, for a FLP™, through switches A, B, and C. The path through switches A, B, and C has been previously scheduled and no header processing is necessary once data units enter the FLP. The link between Switch A and B has a propagation delay of four time frames, i.e., data units transmitted by Switch A during time frame 1 take until time frame 5 (time frames 2 through 5) to reach the other end of the link between Switch A and Switch B. The data units are automatically switched to the proper output port of Switch B and then forwarded to Switch C during time frame 6, arriving at Switch C after three additional time frames (time frames 7 through 9). All data units are guaranteed to arrive at the end of their FLP at the same predetermined rate at which they entered the FLP.
Each FLP's switching schedule is simple, and repeats every time cycle and/or super cycle. Thus, FLPs, together with the predictability provided by the CTR and pipeline forwarding, eliminate the complexity of data packet header processing. Each FLP™ transports data packets of one protocol, such as IP, MPLS, ATM, FR, or FC. However, different FLPs may carry data units of different protocols, i.e., data units of different protocols can travel across a selected link.
Fractional lambda switches have significantly lower complexity than both packet switches and circuit switches with comparable switching capability for the following reasons:                1. Minimum switch fabric complexity is implemented using a Banyan network, which has the complexity of a·N·1gaN switching elements, where N is the total number of optical channels and ‘a’ is the size (number of inputs/outputs) of each switching element.        2. Optimal speed-up of the switch fabric that operates at the same speed as the optical channels (e.g., 10 Gb/s with OC-192 links).        3. Optimal memory access bandwidth is equal to the optical channel bandwidth—the switch architecture enables that, with only 3 input queues, a queue is never used for reading and writing at the same time, i.e., memory access with a speedup of 1.        4. (Very) small input memory for each optical channel (e.g., a 10 Gb/s channel requires 3 input queues with a total of 48 Kbytes of memory) and no buffering at the output port are needed.        5. (Very) simple control of the switch fabric suffices, since its configuration changes at a relatively low frequency (e.g., 80,000 times per second) and it is known in advance. This operation complexity is comparable to that of a T1 multiplexer.        
Though highly efficient, a Banyan Network is subject to what is known as switch blocking: it may be impossible to connect an idle input with an idle output because a switching element is not available on the path between input and output. An interesting attribute of fractional lambda switching is the nearly complete elimination of blocking through Banyan-based switches.
Advances in optical transport have led to the realization of high-speed optical channels (e.g., OC-192 at 10 Gb/s); however, a single source transmitting to a single destination will not utilize the capacity of such channels. In accordance with the present invention, the following two basic requirements have been formed to enable the full utilization of such high capacity channels: (i) Grooming and degrooming: the need to aggregate (i.e., grooming) traffic from multiple sources into one optical channel and to separate (i.e., degrooming) an optical channel traffic to different destinations; (ii) Dynamic optical switching: the need to route portions from one optical channel (i.e., a lambda or a wavelength) on different optical paths to different destinations.
Dynamic all-optical switching is appealing for a number of reasons, stemming from the transparency of the transported data stream to the switching system itself; such reasons being: (i) intrinsically protocol independent (multi-protocol) transport; (ii) high scalability, since the transmission rate of each optical channel is transparent to the optical switching system; and (iii) no processing performed on switched data units, thus eliminating processing bottlenecks. Dynamic all-optical switching is possible when the optical switch reconfiguration time is significantly smaller than the time between two successive switch configuration changes.
The latest advances in optical switching have resulted in decreasing reconfiguration times of optical switch fabrics. However, taking full advantage of such advances for dynamic optical switching is not obvious, for several reasons: (i) Processing of in band control information, e.g., packet headers, is not possible; (ii) Dynamic optical storage is not available to assist in coping with output port contention and switch control and reconfiguration time; (iii) Optical switch reconfiguration time should be significantly smaller than the time between two successive reconfigurations.
Due to the above limitations it is not possible to realize an asynchronous packet switching system, and therefore, using time is necessary. However, time-based techniques deployed in circuit switching (e.g., SONET, based on byte switching, i.e., byte de-multiplexing and byte multiplexing) are not applicable to all-optical switches.
The most comprehensive solution to the above-mentioned problems is use of a common time reference (CTR™) for pipeline forwarding (PF™) in order to facilitate dynamic all-optical switching. CTR™ provides the synchronization needed to orchestrate the control of network switches while eliminating the need for optical storage and processing.
Dynamic All-optical Switching of Time Frames
Time is divided into time frames and any time frame of a sequence of incoming time frames over one optical channel can be optically switched to any outgoing optical channel. Such time frame switching is the basis of fractional lambda switching (FLSw). FLSw is used for constructing Fractional lambda pipes (FLPs), i.e., fractions of a wavelength. Each FLP™ transports data units of one protocol—such as, IP, MPLS, ATM, FR, FC, and SONET frames (e.g., STS1 frame)—, thereby realizing the desired protocol independent property of all-optical switching.
An all-optical switch realizes PF™ in two operational phases. Data units belonging to a whole time frame received from each of the optical channels during the first phase, Phase 1, are switched through the switch in the second phase, Phase 2. In a possible embodiment, if Phase 1 begins in time frame t, Phase 2 takes place in time frame t+1. In another embodiment, if Phase 1 ends in time frame t, Phase 2 takes place in time frame t+1. The 2 phase operation ensures that data units received from the various optical channels are aligned with the CTR before being switched. Phase 2 can be performed during either the time frame immediately following Phase 1, during time frame t+1—immediate forwarding operation, or at a later time frame—non-immediate forwarding operation.
Alignment involves aligning the beginning and end of each time frame from each optical channel with the beginning and end of the CTR™ time frames. Alignment can be performed either independently on each one of the channels carried by an optical link, or collectively on all of the channels carried by an optical link.
Alignment is needed because the propagation delay on optical links between switches is not an integer multiple of time frames. An optical alignment subsystem is part of the all-optical fractional lambda switch, operates on all the wavelengths carried by each optical fiber, and contributes to the realization of Phase 1 of the PF™. The optical alignment subsystem is based on a programmable optical delay line guaranteeing that the overall delay experienced through the optical fiber and the delay line is an integer number of time frames. Data units leave the switch at the transmitting end of a selected optical fiber aligned with the CTR™. When they arrive at the output of the optical alignment subsystem at the receiving end of the selected optical fiber, they are still aligned with the CTR™. The alignment system is comprised of a controller that detects time frame delimiters and adjusts the delay by using a programmable optical delay line (note: the alignment changes only when the propagation delay on the optical link changes).
The availability of a common time reference (CTR™) on a global scale for all network nodes enables the implementation of dynamic optical switches with a simple architecture. This architecture is used to realize fractional lambda switching (FLSw). FLSw is based on a common time reference (CTR™) for pipeline forwarding (PF™) of time frames. In FLSw, the synchronization provided by the CTR is leveraged to orchestrate the operation of subsystems within a switch and across the whole network.